.webp&w=48&q=75){kind=small}
The AI Authority Playbook is a practitioner's guide to earning mentions, citations, and trust inside ChatGPT, Perplexity, Google AI Overviews, Gemini, and Claude.
By Sanjay Bhattacharya
Most of what gets published about “AI SEO” right now is either recycled SEO advice with the word “AI” inserted, or breathless hype about a gold rush that doesn’t hold up once you look at the actual data. This guide is my attempt to do neither. I run an SEO and digital marketing consultancy, I’ve been doing this work for over 16 years, and over the last year AI search visibility has gone from a nice-to-have line item to something nearly every client asks about in the first call.
Here’s the claim I want to make and defend with data: getting cited by ChatGPT, Perplexity, or Google’s AI Overviews is not the goal. It’s a byproduct of building genuine topical authority, structuring content so machines can parse it accurately, and maintaining a consistent entity presence across the web. Brands that chase citations directly, through tactics like stuffing in statistics or adding an llms.txt file because a blog post told them to, tend to see short-lived bumps that don’t compound. Brands that build authority first tend to show up everywhere, including in AI answers, without specifically optimizing for any one engine.
Who this is for: marketing leads, founders, and in-house SEO teams who already understand traditional search and want a grounded, sourced view of how AI search actually works right now, what’s confirmed by research versus what’s still speculation, and what to actually do about it. I’m writing this from the accounts we manage, not from a vendor’s sales deck, which means I’ll tell you where the data is thin or contradictory rather than smoothing it over.
What you’ll find here: a sourced breakdown of the current state of AI search, including the uncomfortable fact that AI engines and Google increasingly disagree about which pages deserve to be cited; a framework for thinking about AI visibility as an output of authority rather than a target in itself; the AI Authority Model, the AI Recommendation Journey, and three other original frameworks I use with clients; a tool-by-use-case breakdown of the AI visibility monitoring market as it actually stands today, with real pricing; decision trees for build-vs-buy and platform selection; anonymized results from real engagements where AI search work was layered into active SEO programs; and a 90-day roadmap you can actually execute.
“Citations are an output, not the objective.” That single distinction is the difference between a GEO program that compounds and one that produces a brief, hollow spike.
For years, the working assumption in SEO was simple: rank well on Google, and you’d eventually show up everywhere else that mattered, since most other engines leaned on Google’s index in one form or another. That assumption is breaking down quickly. Multiple research firms have now measured the overlap between Google’s top-10 organic results and the sources that AI engines actually cite, and the trend line is consistent even if the exact number varies by methodology: that overlap has fallen from somewhere around 70-75% in mid-2025 to a range as low as 17-38% in early 2026, depending on which study you read.
That’s not a small drift. It means a page can rank #1 on Google for a query and never appear once when someone asks ChatGPT or Perplexity the same question, and vice versa. Ahrefs’ research backs this up from another angle: roughly 28% of ChatGPT’s most-cited pages have zero organic visibility in Google at all. These are pages Google barely acknowledges that AI engines treat as authoritative sources.
There’s an important nuance here that most GEO content glosses over, and it matters for how you allocate effort. For most of 2025, Google’s own AI Overviews behaved very differently from open engines like ChatGPT and Perplexity, because AI Overviews were grounded directly in Google’s existing search index and the overlap with traditional top-10 rankings stayed high, with analyses through mid-2025 putting it above 90%. That’s no longer the full picture, and the shift is one of the most important updates to this guide since it first published. Ahrefs’ large-scale study of 863,000 keywords and 4 million AI Overview URLs found that the share of AI Overview citations coming from top-10 organic pages fell from 76% in July 2025 to 38% by March 2026, and a separate BrightEdge analysis using a different methodology puts the current overlap even lower, around 17%. The timing lines up closely with Google’s global upgrade of AI Overviews to Gemini 3 in late January 2026. The likely mechanical explanation ties directly into query fan-out, covered next: AI Overviews increasingly draw citations from the sub-query SERPs generated during fan-out, not just the primary top-10 results, so a page can rank #1 for the head term and still lose the citation to a page that answers one of the fan-out sub-questions more completely. Ranking well still helps. Position 1 still carries meaningfully better citation odds than position 10. But strong traditional SEO alone is no longer close to sufficient for AI Overviews specifically, and treating AI Overviews as “basically covered by good SEO” is a riskier assumption now than it was a year ago. Treating “AI search” as one undifferentiated channel, when it’s really at least two different retrieval systems with different and now-converging rules, is one of the most common strategic mistakes I see.
The mechanical reason for the gap comes down to how these systems actually retrieve information. When someone asks an AI engine a question, the system typically doesn’t run that exact phrase as a single search. It breaks the question into several related sub-queries, runs each one separately, and then synthesizes an answer from whichever sources best answer each piece. Google’s own developer documentation describes this as query fan-out and gives a clean example: a query about fixing a weedy lawn might get split into searches for the best herbicides, chemical-free removal methods, and prevention techniques, each pulling from different sources.
The practical implication is that a single excellent page targeting one keyword phrase isn’t enough anymore. AI engines reward brands that have built out content addressing the full cluster of sub-questions around a topic, because a brand with five pages covering a topic from different angles gets pulled into more of those fan-out sub-queries than a brand with one strong page. This is the single biggest mindset shift I’d ask you to take from this section: stop auditing content by individual keyword, and start auditing it by topic cluster, mapping every sub-question a buyer might ask and checking whether you have a direct answer to each one.
ChatGPT crossed 1 billion weekly active users in August 2026, according to OpenAI, up from 900 million in February 2026 and 400 million a year before that. At Google I/O in May 2026, Google folded AI Mode into one integrated experience with AI Overviews and reported AI Overviews reaching more than 2.5 billion monthly users, with AI Mode itself surpassing 1 billion monthly active users roughly a year after launch. Google AI Overviews now appear on a meaningful share of informational searches, with tracking through the first half of 2026 putting the figure around 48% for the industries Google triggers them most aggressively in. Zero-click search behavior, where a user gets their answer without clicking through to any website, has kept climbing in parallel: multiple 2026 studies put the share of all Google searches ending without a click in the 43-60% range, rising to roughly 83-93% specifically when AI Overviews or AI Mode appear.
What makes this worth real budget rather than experimental budget is the conversion data. Multiple independent studies, run by different firms using different methodologies, all point in the same direction even though the exact multiples differ: traffic referred from AI engines converts meaningfully better than traditional organic search traffic. Estimates range widely depending on the source and industry, from roughly 3x better conversion in aggregated agency data up to 20x-plus in some platform-specific analyses, with the consistent explanation being that a user arriving from an AI citation has already received an implicit endorsement and detailed context before they ever land on your site. I’d treat any single multiple you read with some skepticism since the studies aren’t standardized, but the direction of the effect shows up too consistently across too many independent sources to dismiss.
In keeping with how I approach every client recommendation, I want to separate what current research has actually demonstrated from what the GEO industry is currently guessing at, because a lot of published advice blurs this line.
Reasonably well confirmed: The original Princeton, Georgia Tech, and IIT Delhi study presented at KDD 2024, which coined the term Generative Engine Optimization, ran controlled tests across 10,000 queries and found that specific content interventions measurably increased citation visibility. Adding direct quotations from credible sources produced roughly a 41% lift, adding statistics produced roughly 32%, adding citations to outside sources produced roughly 30%, and improving general fluency and readability produced roughly 28%. This is peer-reviewed, methodologically transparent research, not a vendor claim, and it’s the closest thing this field has to ground truth.
Reasonably well confirmed: Schema markup and structured data measurably improve machine readability and discoverability. Independent analyses across hundreds of brands found that missing structured data elements directly costs AI visibility coverage, and one large-scale study found sites present consistently across four or more platforms were nearly three times more likely to appear in ChatGPT recommendations than sites with thin or inconsistent presence.
Still mostly inference: the exact internal weighting any specific AI engine gives to factors like domain age, backlink profile, or content recency. Vendors in this space publish a lot of confident-sounding percentages about ranking factors, but unlike traditional Google SEO, none of these AI labs have published an actual ranking algorithm, and most of what’s circulating as “the formula” is reverse-engineered from limited sampling. I’ll flag these clearly wherever they come up in this guide rather than presenting industry consensus as settled fact.
Still mostly inference: precise citation freshness windows. It’s well established directionally that AI engines favor recently updated content and that citations to a page drop off as it ages, but the specific timelines you’ll see quoted, such as a particular page losing relevance after exactly three months, come from individual vendors’ proprietary tracking rather than from independently replicated research, and I treat them as a useful planning heuristic rather than a hard rule.
AI search isn’t a separate discipline you bolt onto SEO. It’s evidence that the underlying job, building genuine, well-structured, frequently updated topical authority, has gotten more demanding and more rewarding at the same time. The brands showing up consistently in AI answers right now aren’t running secret AI hacks. They’re the ones who already did the unglamorous work of structured data, clear writing, topic cluster coverage, and consistent entity presence across the web, and who are now seeing that work pay off in a second channel they didn’t have to build from scratch.
Semrush's AI Visibility Index 2026, drawn from 126 million real US AI search prompts across ChatGPT, Google Gemini, Google AI Mode, and Google AI Overviews, makes a distinction that's easy to miss and important to track separately: a mention (your brand named in an AI answer) and a citation (your domain quoted as a source) are earned through different mechanisms and don't reliably move together. Mentions come from brand authority and category fit. Citations come from content depth. A brand can be mentioned constantly and cited almost nowhere, or the reverse.
The clearest illustration is Wikipedia, which is cited constantly across every AI platform but is rarely the brand an AI answer is actually discussing. The inverse shows up too: brands like Patagonia carry heavy mention volume, built through third-party gear-review coverage, relative to their own-domain citation footprint. Across the four platforms Semrush tracked, the share of mentioned brands that are also cited as sources ranges from as low as 30% on Gemini up to 64% on Google AI Overviews. That gap alone tells you these are two different things to build for, not one.
The practical implication for the rest of this guide: when Section 8.3 talks about tracking “citation share,” that's now two separate numbers to log per platform, not one blended metric. A brand showing up in AI answers without picking up citations still needs the Layer 3 and Layer 4 work described in Section 4. A brand collecting citations without mentions, more common for reference and comparison sites than for a typical client brand, has a different problem: it's infrastructure for someone else's answer, not a recognized entity in its own right.
I want to spend a full section on this because it’s the single most common strategic error I see brands make once they decide AI search visibility matters. The instinct, understandably, is to treat “get cited by ChatGPT” as a KPI you optimize directly, the same way you’d optimize for a keyword ranking. That instinct is wrong, and acting on it tends to produce work that looks productive without actually building anything durable.
When a team treats AI citations as the target, the work that gets prioritized tends to be shallow by design: adding a statistic here, inserting a quote there, generating an llms.txt file because a blog post said AI crawlers want one. Some of these tactics have measurable short-term effect, per the Princeton research cited above. But none of them address why an AI engine would trust your brand as a source in the first place. Google’s own developer guidance on this point is direct: their generative AI features are rooted in the same core ranking and quality systems as traditional search, and they explicitly tell site owners to ignore “AEO/GEO hacks” like artificial content chunking or unnecessary AI text files in favor of fundamentally sound SEO practice.
There’s also a structural reason citation-chasing doesn’t compound: citation share is volatile in a way keyword rankings never were. A brand can lose a citation it held for months because a competitor published a more recent statistic, or because the underlying model was updated and now weights a different signal slightly differently. Industry monitoring data suggests that when a brand loses ground in AI answers, a competitor displacing that citation is the cause in the large majority of cases. If your whole strategy is built around holding specific citations, you’re defending sandcastles. If your strategy is built around being the most trustworthy, most complete, most consistently updated source on a topic, the specific citations come and go but the underlying authority holds.
A recent Semrush survey of 481 marketers on AI search behavior puts a number on this. Teams with fully integrated SEO and AI search execution report meaningfully more traffic or leads from AI platforms, at 81%, versus 36% among teams running the two as separate workflows. The same survey found 45% of marketers can't properly measure their AI visibility at all, and only 9% measure the full set of metrics that matter. Read together with the mentions-versus-citations point in Section 2.6, this looks less like a tooling gap than an organizational one: the brands seeing results are treating AI visibility as one connected discipline with proper measurement, not a side project bolted onto SEO.
This is where I’d point you back to fundamentals that long predate the term GEO. Research from Yext analyzing 6.8 million AI citations found that the large majority, 86%, trace back to sources a brand directly controls: its own website and its business listings and directories. That’s a genuinely useful finding because it means AI visibility, unlike traditional link-driven SEO authority, is something most brands can build primarily through their own owned channels rather than needing to win earned coverage from third parties first.
In the accounts we manage, the work that actually moves AI visibility looks almost identical to the work that builds durable SEO authority: comprehensive, well-structured content that fully answers a topic rather than skimming it; consistent entity information (your name, your services, your locations, your credentials) presented the same way across your site, your Google Business Profile, your directory listings, and any other place your brand appears online; clear authorship and credentials, since AI systems increasingly weigh the experience and expertise signals that Google has been training reviewers to look for under E-E-A-T for years; and a cadence of updates rather than a one-time publish, because both Google’s ranking systems and AI retrieval systems penalize content that visibly hasn’t been touched in a long time.
I tell clients to stop asking “how do we get cited by ChatGPT” and start asking “what would make us the obvious source on this topic, regardless of which engine someone asks.” The first question leads to tactics. The second question leads to a content and entity strategy that happens to produce citations as a side effect, across every engine simultaneously, without having to reverse-engineer each platform’s retrieval quirks separately. The frameworks below are built entirely around operationalizing that second question.
This is the framework I use to diagnose where a client’s AI visibility problem actually lives, because “we don’t show up in ChatGPT” can mean four completely different things depending on which layer is broken. I think of AI authority as four stacked layers, where each layer depends entirely on the one beneath it. You cannot buy your way into Layer 4 if Layer 1 is broken, no matter how much content or PR budget you throw at it.
Figure 1: The AI Authority Model. Each layer is a prerequisite for the one above it.
This is the layer almost everyone assumes is fine and almost no one actually checks. Foundation means an AI crawler can technically reach, render, and parse your content at all. The most common failure here, and one of the most common problems I see across the accounts we manage, is a robots.txt file that’s silently blocking AI user agents, often because a CDN or security plugin changed its default configuration without anyone noticing. Cloudflare’s default settings shifted toward blocking AI bots automatically in 2025, which means a meaningful number of sites lost AI crawler access without their team doing anything at all, and this is worth re-checking now rather than treating it as solved. As of July 2026, Cloudflare replaced the blanket “Block AI Bots” switch with three separately controllable categories, Search, Agent, and Training, available to every customer including the free tier. Starting September 15, 2026, Cloudflare will block mixed-use crawlers, ones that blend search, agent, and training functions, by default on any page carrying ads, for new domains and any site that already has AI blocking turned on. The practical wrinkle: Cloudflare has named Googlebot, Applebot, and Bingbot as mixed-purpose crawlers under this rule, so a site with AI blocking already enabled could inadvertently catch Googlebot on ad-supported pages unless someone explicitly opts it out before the deadline. Checking server logs for user agents like ChatGPT-User, ClaudeBot, and PerplexityBot, confirming your robots.txt explicitly allows the ones you want reaching you, and reviewing your Cloudflare crawler categories against this September deadline specifically, is the first thing I do on any new AI visibility engagement, before touching a single piece of content.
Once a page is reachable, the question becomes whether an AI system can cleanly extract meaning from it. This is schema markup (Article, FAQPage, Organization, and where relevant ItemList structured data), clear heading hierarchies, and what I’d call definition-lead writing, where every major section opens with a self-contained sentence that answers the implicit question a reader, or a model, brought to that section. Independent analysis of AI citation data has found that missing structured data elements measurably costs visibility coverage, and that schema markup improves machine discoverability by a wide margin, though I’d stop well short of treating schema alone as sufficient. It’s a multiplier on substance, not a replacement for it.
This is where most of the real work lives and where most shortcuts fail. Substance means genuine topical depth across a full cluster of sub-questions, original data or first-hand observations rather than restated consensus, and clear authorship with real credentials attached. This maps directly onto Google’s E-E-A-T framework, and there’s good reason to believe AI systems weigh similar signals, since both are ultimately trying to answer the same underlying question: should this source be trusted? The Princeton KDD 2024 research is most directly relevant here. Quotations, statistics, and citations to outside sources all measurably increase visibility, and all three are substance-layer interventions, not structure-layer ones. You can’t schema-markup your way to a 41% citation lift. You earn it by actually including a credible quotation.
The top layer is everything that happens off your own site: directory and listing consistency, third-party mentions, brand search volume, and presence across the platforms AI systems draw on when forming a picture of your brand. Yext’s analysis of 6.8 million citations found 86% trace back to brand-owned sources, primarily first-party websites and business listings, which is genuinely good news since it means most of Layer 4 is still within a brand’s direct control rather than dependent on earning third-party press. One large cross-platform study found brands with a consistent presence across four or more platforms were nearly three times more likely to appear in ChatGPT recommendations than brands with thin, inconsistent presence.
In practice, when a client says “we’re invisible in AI search,” the honest first step is figuring out which layer is actually broken, not assuming it’s Layer 3 or 4 when it’s sometimes a robots.txt file.
The Authority Model explains what to build. This framework explains what actually happens between the moment a user types a question and the moment your brand either does or doesn’t show up in the answer. I’ve mapped it as six steps, and I’ve deliberately color-coded which steps are documented, observable behavior versus which step is still genuinely a black box, because conflating the two is where a lot of GEO advice goes wrong.
Figure 2: The AI Recommendation Journey, from query to conversion.
A user submits a natural-language query. The model performs query fan-out, breaking that query into several related sub-queries, a behavior Google’s own developer documentation describes explicitly and gives concrete examples of. Each sub-query then triggers retrieval, where a retrieval-augmented generation system pulls candidate source documents that appear semantically relevant. This is mechanical and reasonably well understood, and it’s the basis for the topic-cluster content strategy I’d recommend over single-keyword pages.
Synthesis is where the model drafts an answer from the retrieved candidates and, critically, decides which sources are trustworthy enough to cite by name. This is the step every GEO vendor claims to have cracked, and it’s also the step where I’d ask you to be most skeptical of confident percentages. No major AI lab has published the actual weighting logic here. What we have is correlational research, like the Princeton study’s finding that quotations, statistics, and citations correlate with higher visibility, which is genuinely useful, but it’s describing what correlates with being cited, not a confirmed causal ranking formula. I treat Step 4 as the frontier of what’s knowable right now, and I’d be cautious of anyone who tells you otherwise with total confidence.
If your brand clears Step 4, you get named or linked in the synthesized answer. What happens next is the part that actually justifies the investment: the conversion data referenced in Section 2 consistently shows AI-referred visitors converting at meaningfully higher rates than traditional organic visitors, because by the time someone clicks through from an AI citation, they’ve already received an implicit endorsement and arrive with real context about why your brand might fit their need.
The AI visibility tooling market has moved fast, and it raised more than $300 million in funding between mid-2025 and early 2026 alone. That speed has produced a genuinely confusing landscape, where the question isn’t “which tool is best” but “which category of tool actually solves the problem in front of you.” Here’s how I break it down for clients, organized by what each category is actually for rather than by brand name.
These are purpose-built tools that run prompts on a schedule across multiple AI engines, parse the responses, and track citation share, sentiment, and competitive position over time. This is the deepest category for teams that treat AI visibility as a core, ongoing discipline rather than a side metric. The category has also started consolidating: Profound closed a $96 million Series C in February 2026 at a reported $1 billion valuation, and Sitecore acquired Scrunch AI in June 2026 to fold it into its digital experience platform roadmap. Expect more of this. When a category raises $300 million-plus in under a year, as this one has, some of these vendors will get bought, some will get folded into SEO suites, and pricing pages will keep moving. Treat the specific numbers below as directionally right rather than exact at the moment you read this, and check current pricing before you commit budget.
| Platform | Best For | Starting Price | Key Limitation |
| Profound | Enterprise teams needing the deepest citation database and brand-perception analysis | $99/mo (ChatGPT-only tier); broader coverage from $399/mo; enterprise custom | Built for teams that already know what to do with the data; expensive if no one owns execution |
| Peec AI | Mid-market teams wanting simple, flexible, multi-market prompt tracking | From roughly $89-199/mo (Pro tier) | No content creation or optimization tooling, monitoring only |
| Otterly.ai | Smaller teams or agencies wanting an accessible entry point | From roughly $29-79/mo | Less depth on brand-perception and narrative analysis than enterprise tools |
| Scrunch AI | Enterprises that also need to see what AI crawlers encounter on-site (not just output-side citations) | From $250/mo since Sitecore's June 2026 acquisition; Enterprise custom | Acquired by Sitecore in June 2026; roadmap is now tied to Sitecore's DXP suite rather than a standalone product |
If you’re already paying for Semrush or Ahrefs, both now offer AI visibility tracking as an add-on inside the same workspace. The advantage is zero new vendor relationship and immediate access to your existing keyword and backlink data alongside AI metrics. The trade-off is real and worth understanding before you rely on these as your primary source of truth.
| Platform | Best For | Starting Price | Key Limitation |
| Semrush AI Visibility Toolkit | Teams already in Semrush wanting AI tracking layered onto existing SEO workflows | From $99/mo add-on; bundled tiers $199-549/mo | Global prompt index has been shown to surface off-topic mentions that distort the visibility score; verify before trusting it for competitive claims |
| Ahrefs Brand Radar | Ahrefs users wanting AI share-of-voice tied to existing search-demand data | $199/mo per AI index, or $699/mo for all six | Does not track Claude or Grok; full bundle plus base Ahrefs subscription gets expensive fast |
A worth-noting data point on that Semrush limitation: independent testing comparing two real consumer banking brands found Semrush’s global index showing the smaller brand with double the AI mentions of the clear market leader, and on investigation, the bulk of those mentions traced to off-topic queries (sponsorships, unrelated cultural references) rather than the brand’s actual core business. The lesson isn’t that Semrush is bad. It’s that any single tool’s headline visibility score deserves a sanity check against what’s actually driving it before you report it to a client or a board.
The build-vs-buy question in AI visibility is not really about software. It is about whether your team has the capacity to act on what a tool tells you, and whether the cost of a tool is justified by the decision quality it produces. I have seen teams buy enterprise monitoring platforms and never move on the data. I have also seen scrappy in-house teams run a fully defensible AI visibility program out of a shared spreadsheet for six months while they validated whether the channel was worth more investment. Both can be the right answer depending on the situation.
The most useful question before evaluating any tool is: what decision will this data unlock? If the answer is “we need to know whether AI search is a real source of business for us yet,” you do not need a $400/month platform. A two-hour manual audit, running your 10-15 highest-value queries across ChatGPT, Perplexity, and Google AI Overviews by hand and logging results in a spreadsheet, gives you a defensible baseline for under $50 in staff time. If the answer is “we manage AI visibility across 12 client accounts and need automated, engine-specific citation tracking with competitive benchmarking,” a dedicated platform pays for itself almost immediately in analyst hours saved.
These four frameworks represent the most common decision points I walk clients through when starting an AI visibility program. I have simplified each into a repeatable logic flow because the questions come up in almost every engagement, and having a consistent framework prevents teams from reliking answers based on bias toward the tactics they already know.
The answer depends almost entirely on how you acquire customers today. If more than 60-70% of your organic traffic comes from Google and your target audience skews toward traditional search behavior (local services, healthcare, legal, e-commerce with high-purchase-intent searches), your AI Overviews strategy is almost entirely served by strong traditional SEO. The citation overlap between traditional top-10 rankings and AI Overviews is high enough that you rarely need a separate workstream. Invest in open-engine visibility (ChatGPT, Perplexity, Gemini) when your audience is demonstrably using those platforms to research your category, which is most true for software, B2B services, financial products, and complex consumer decisions where people ask long, nuanced questions.
Even within “open engines,” treating ChatGPT, Gemini, and Google AI Mode as one bucket is a step too coarse. Semrush's four-platform breakdown of the same US prompt database shows each one leaning on a different source diet. ChatGPT cites Reddit and Wikipedia most heavily and pulls from the most sources per answer, around 15 per response. Gemini stays close to Google's own commerce and shopping surfaces and cites the fewest sources of the four, around 3. Google AI Mode over-indexes on social platforms and, notably, local-discovery brands like Yelp. AI Overviews leans hardest on YouTube and established reference sites, driven by the fact that its prompts average roughly 25 characters versus 55-58 characters on the other three, because they're really repurposed Google Search keywords rather than conversational questions. For a local-services or single-location client, AI Overviews is often the highest-leverage platform to chase, since it rewards the same short, definitional, keyword-led content that traditional local SEO already produces. For B2B, software, or complex-decision categories, ChatGPT's community-and-reference weighting makes Reddit presence and Wikipedia accuracy disproportionately valuable in a way that doesn't transfer cleanly to Gemini or AI Mode.
Map your buyer’s highest-intent questions, the queries someone asks when they are genuinely close to a purchase or service decision, and audit whether you have a direct, self-contained answer to each one anywhere on your site. Priority goes to: any high-intent question that you have no published answer to (content gap); any question where you have content but it is more than 12 months old with no update (freshness gap); and any question where your existing content answers the topic but is buried in long-form prose without a clear definition-first opening sentence (structural gap). These three gap types map directly onto the Structure and Substance layers in the Authority Model.
This is the measurement question the industry is still genuinely working out, and I want to be direct that there is no clean GA4 channel breakdown for “AI search traffic” yet. What you can track today: referral traffic from known AI engine domains (perplexity.ai, chat.openai.com, gemini.google.com, and others), which appear in your analytics as direct or referral depending on the user’s browser and whether the engine passes a referrer header; branded search volume trend in Google Search Console, which tends to rise when AI citations increase brand discovery; citation share tracked manually or via a platform tool; and conversion rate of visitors from AI-engine domains versus organic average, which is the metric that tends to make the business case most compellingly when you have enough volume to measure it.
One addition to the tracking list above, per Section 2.6: log mentions and citations as two separate columns per platform, not one blended “AI visibility” number. A single aggregate score can hide which platform is actually driving movement, and it can make a brand that's heavily mentioned but rarely cited, or the reverse, look flat when it's actually strong on one dimension and weak on the other. Track both, per platform, and the diagnosis gets much easier.
Almost always start with existing content. The most common finding in any AI visibility audit is not that you lack content on your core topics but that the content you have is not structured for machine extraction. A page that fully covers a topic but opens with a branding paragraph, buries the key answer in paragraph four, and has no schema markup will consistently underperform a shorter, less comprehensive page that opens with a definition-first sentence, uses clear H-tag hierarchy, and has FAQ schema attached. The optimization sequence I follow: identify your 20 most commercially important topic pages, run each through a structure audit (does it open with a direct answer? are key questions called out as headings? is FAQ schema applied where relevant?), fix those structural issues first, and only then evaluate whether the substance is thin enough to warrant additional content production.
The four frameworks in this section are my own practitioner models, developed and refined through the AI visibility work we do at Sanjay B. Consulting. They are not academic constructs and they are not vendor-sponsored. They are the mental models I actually use when diagnosing a client’s AI search situation and designing the work plan. I am sharing them here because the most useful thing a practitioner guide can do is make its reasoning transparent, not just its conclusions.
Figure 3: The AI Authority Flywheel. Compounding visibility is the mechanism; citation is one point in the loop, not the goal.
The flywheel is my answer to the “why bother with AI authority when citation share is so volatile?” objection. Individual citations come and go, but the underlying brand entity strength that earns them does compound if the flywheel is turning. The loop works like this: you publish structured, credentialled content (Publish); AI crawlers discover, parse, and cache it (Index); your brand appears in an AI answer for a relevant query (Cite, shown in red because it is the visible output of the loop, not its engine); a pre-qualified visitor arrives at your site with context and intent (Attract); meaningful engagement signals, time on site, low bounce, repeat visits, send positive behavioral signals back to both traditional and AI-engine evaluation systems (Amplify); which increases your brand entity strength and makes the next cycle’s Publish step more likely to result in a citation. The compounding happens gradually. Nothing in this loop works after one rotation.
Figure 4: The AI Trust Layers. Trustworthiness is triangulated outward from entity clarity at the core.
This framework explains how AI engines appear to assess whether a brand is trustworthy enough to cite, working outward from the most controllable signal (your own entity definition) to the least controllable (ecosystem signals from third parties). The Core is entity clarity: your brand name, category, primary services, and location presented identically across every surface you control. Inconsistency at the core, say, a slightly different business name on your Google Business Profile versus your website versus your Yelp listing, creates ambiguity that downstream trust signals cannot fully compensate for. The Inner layer covers on-site authority: author credentials, E-E-A-T implementation, structured data, and content update cadence. The Mid layer is corroboration: the degree to which your on-site claims are confirmed by third-party directories, external citations, and structured knowledge sources like Wikipedia or Wikidata. The Outer layer is ecosystem signals: brand search volume, PR mentions, social proof, and aggregated third-party reviews. AI engines cross-reference all three outer layers against your core. Gaps in any layer introduce doubt. Consistency across all layers compounds trust.
Figure 5: The AI Search Maturity Model. Five stages from invisible to authoritative; most mid-market brands start at Stage 2.
Most diagnostic conversations about AI visibility get stuck because the team and the consultant are implicitly describing different problems. One person is worried about why they do not appear in ChatGPT at all. Another is trying to understand why they appear for branded queries but not for high-intent category queries. A third wants to know why a competitor is the default recommendation even when your product is technically superior. These are three different maturity stages with three completely different solutions. The model gives us a shared vocabulary. Stage 1 (Invisible) is a technical access and entity definition problem. Stage 2 (Discoverable) means AI crawlers can reach you but have not found enough depth to cite you consistently. Stage 3 (Cited) means you appear for branded and close-in queries. Stage 4 (Recommended) means you are being cited unprompted in competitive category comparisons. Stage 5 (Authoritative) is when competitors’ content starts citing you as a source, which is the compounded authority the flywheel is designed to build toward. In my experience, most mid-market brands coming to this work for the first time sit at Stage 2. The jump to Stage 3 is primarily a content-depth and structured-data investment. Stage 3 to Stage 4 requires sustained cadence and the off-site entity work described in the Trust Layers framework. Stage 5 is a 12-24 month program, not a campaign.
One useful calibration on top of the Maturity Model: how much ambition is realistic depends on how concentrated a category already is. Semrush's category-level analysis of the same dataset found the share of top-10 AI mentions held by the top three brands varies by about 41 percentage points across verticals, from News & Media at roughly 83% down to Finance at roughly 41%. In a concentrated category, a Stage 3 or 4 brand should treat position 4-10 as a realistic target rather than expecting to unseat the top three; in a flat category, focused work can move a brand's rank within a single quarter. I'd check where a client's category sits on that curve before setting a Stage 4 target, so the roadmap in Section 11 is calibrated to what's actually achievable rather than to a generic ambition level.
This is the operational framework I use for ongoing content management once the foundational work is done. It has four steps that run on a monthly or quarterly cadence depending on the pace of change in your category. First, Monitor: run your tracked prompt set across engines and log citation share, new citations, and lost citations with their apparent reasons. Second, Diagnose: for each citation loss, identify which layer of the Authority Model is failing (structure issue? freshness issue? a competitor published something more complete?). Third, Produce: create or update the content that closes the identified gap, prioritizing freshness updates and structural fixes over net-new content creation since those are faster and often more impactful. Fourth, Validate: rerun the relevant prompts 3-4 weeks after the update and check whether the gap has closed. The Loop is deliberately simple because the discipline of running it consistently matters far more than the sophistication of the tooling around it. I have seen teams with enterprise AI monitoring platforms that never complete steps 2-4. The Loop does not work if you stop at step 1.
The following four case studies are drawn from real client engagements at Sanjay B. Consulting. All clients are anonymized under NDA, identified only by industry and geography. In each case, AI search visibility was integrated into an active SEO program rather than run as an isolated discipline, and the results presented reflect that integrated program. Where AI-specific metrics (such as AI answer presence percentage) were directly measured, I present them as measured. Where AI visibility was a component of overall growth but cannot be cleanly separated from broader SEO gains in the underlying analytics, I say so explicitly.
| Metric | Before | After | Change |
| SERP Positions | Baseline | +83.7% | More positions |
| Organic Traffic | Baseline | +55.5% | Traffic growth |
| Referring Domains | 300 | 498 | +86.3% |
| AI Answer Presence | Not measured | 50-60% | Target achieved |
A faith community organization in South Miami with a nearly century-long community presence had an online footprint that did not reflect its standing. Despite being a recognized institution, the organization had minimal organic visibility, a low-traffic website, and no structured approach to AI search. The work began with a full technical audit that identified crawlability gaps, weak keyword targeting, and content that was neither optimized for search nor structured for machine extraction. On-page optimization, off-page directory and branding work, and AI-focused structured data implementation ran in parallel.
The AI-specific work involved applying schema markup and entity optimization to achieve consistent, measurable AI answer presence across local faith community queries. By month seven, the organization was appearing in AI-generated answers for South Miami faith community searches at a 50-60% rate, a metric that was directly tracked through manual prompt monitoring across Google AI Overviews and ChatGPT. The 83.7% increase in SERP positions and 55.5% traffic growth reflect the integrated program. The AI visibility component specifically was the piece that drove new congregation members who discovered the organization through an AI-generated answer rather than a traditional search result.
The lesson: for a brand with strong underlying authority and a weak technical foundation, structured data and entity optimization can unlock AI visibility relatively quickly once the Foundation layer is fixed.
| Metric | Before | After | Change |
| Total Users | 224 | 1,728 | +671.4% |
| Organic Leads (Key Events) | 0 | 192 | +99% (new) |
| Backlinks | 478 | 2,034 | +325.5% |
| Organic Keywords | 1,930 | 2,158 | +11.6% |
A veterinary practice in Northern California was virtually invisible online despite being a trusted local clinic. With competitors embedded in local search results and a Google Business Profile that was significantly underperforming, the practice was losing patients to less established competitors who had invested in digital presence. The 90-day program concentrated on the Foundation and Structure layers almost exclusively: technical SEO fixes, on-page overhaul of service pages, Google Business Profile build-out including service listings, Q&A content and photo optimization, and geo-specific keyword targeting for Oakdale-area searches.
The 671% growth in total users in 90 days is the figure that tends to stop people in their tracks, and the honest explanation is that it reflects starting from a very low base. The more significant operational result was the creation of 192 net-new organic lead events (appointment request form submissions and call clicks) from zero, because before the engagement those conversion events were not even tracked. The roadmap in Section 11 is partly informed by this engagement, specifically the prioritization of Google Business Profile optimization and citation consistency in Month 1 as Foundation-layer work that often produces outsized local results before any content is published.
The lesson: the fastest AI and organic visibility gains tend to come from brands that have real-world authority but a broken technical Foundation. Fixing the Foundation is not glamorous but it is consistently the highest-leverage first move.
| Metric | Before | After | Change |
| Total Users | 1,300 | 5,100 | +260% |
| Organic Traffic | 1,300 | 4,900 | +208% |
| Organic Leads | Baseline | +46.1% | Increase |
| Flagship Keyword | Not ranking | #1 | Canadian immigration lawyer |
A Canadian immigration law firm with offices in both Canada and India needed to build credible digital authority simultaneously in two highly competitive markets where the client base had entirely different search behavior. A student in Chandigarh searching for Canadian immigration help asks fundamentally different questions than a permanent resident in Toronto navigating a spousal sponsorship. The work required a dual-region content strategy where each piece of content was built around the specific search intent of one market rather than trying to serve both at once, with separate internal architecture for India-facing and Canada-facing content.
The AI search visibility component was a deliberate part of the brief from the start: the firm wanted to appear in AI-generated answers on ChatGPT, Perplexity, and Google AI Overviews for immigration queries, because that is increasingly where clients in both markets research complex legal decisions before contacting a firm. Structured content, schema markup, and entity authority work were layered onto the content strategy specifically to achieve this. By month six the firm was ranking #1 for its flagship keyword and appearing in AI-generated immigration answers in both markets. The 260% user growth and 208% organic traffic growth reflect the combined effect of all workstreams. AI search visibility was a component that extended reach to users who were asking AI engines for help, not using Google directly.
The lesson: for high-trust, high-complexity service categories where buyers research extensively before making contact, AI search visibility is a pipeline-expansion tool, not just a brand awareness metric.
| Metric | Before | After | Change |
| Total Users | 5,700 | 11,000 | +92.2% |
| Organic Keywords | 1,700 | 3,300 | +208.6% |
| Flagship Keyword | Position 80 | Position 1 | +79 positions |
| Event Count (Conversions) | 29,500 | 56,300 | +90.5% |
A national platform offering a free prescription discount card had a genuinely valuable product but virtually no organic visibility for the high-intent download queries that would drive activations. The challenge was part SEO and part information architecture: the site needed to be found both by users in specific states searching for local pharmacy savings and by users searching nationally for download-intent terms, which required a complex geo-targeting content architecture built at state-level scale.
This was the longest engagement of the four and the one where the compounding effect of sustained AI visibility work is most visible. The 208% growth in organic keywords was the result of the state-level content architecture going live progressively over months three through nine. The AI search visibility work, building topical authority and structured content to appear in AI-generated answers on ChatGPT and Perplexity when users ask how to save on prescriptions, extended the platform’s reach to the specific audience, uninsured and underinsured Americans researching options, who are most likely to be using AI engines to ask open-ended questions about healthcare costs rather than typing a specific product name into Google. The 90.5% increase in conversion events over 12 months reflects the integrated program. AI-referred traffic was a measurable contributing channel by month six, appearing in analytics as referral traffic from AI engine domains at a conversion rate meaningfully above the organic search average.
The lesson: for platforms serving audiences with urgent, high-anxiety needs (healthcare costs, immigration status, financial decisions), AI search is where the most vulnerable and highest-intent users are asking for help first. Being present in those answers is both a business opportunity and, in categories like this, a genuine service.
This roadmap is designed to take a brand from wherever it currently sits, most likely Stage 2 in the Maturity Model, to Stage 3 in 90 days, meaning you move from “crawlable and indexed” to “appearing in AI answers for branded and close-in queries” with a clear path toward Stage 4 afterward. It is built around the constraint that most teams have: limited bandwidth, so every task needs to be sequenced in order of leverage rather than comprehensiveness.
Figure 6: The 90-Day AI Authority Roadmap. Foundation tasks in Month 1 are prerequisites; do not compress or skip them.
Week 1-2: Crawl and access audit. Check server logs for AI user agents (ChatGPT-User, PerplexityBot, ClaudeBot, Google-Extended). Verify robots.txt explicitly permits all AI crawlers you want to reach you. Run a full technical crawl to confirm core pages are indexed. Fix any rendering issues on JavaScript-heavy pages that prevent clean content extraction.
Week 2-4: Schema markup on five priority pages. Start with your five most commercially important pages: homepage, primary service or product pages, and any page targeting a high-intent query you already rank for. Apply Organization, Article or WebPage, and FAQPage schema where relevant. This is a fast, high-leverage intervention that measurably improves machine readability before any content is written.
Weeks 1-4: Listing consistency audit. Run your brand name, address, phone, and category information through your five most important directories (Google Business Profile, Bing Places, Apple Maps, Yelp where relevant, and any industry-specific directories). Fix any inconsistencies. This is a Core layer fix in the Trust Layers model and is a prerequisite for the Signal work in Month 3.
Weeks 1-4: Manual baseline tracking. Run your 10-15 most important queries across ChatGPT, Perplexity, and Google AI Overviews. Log whether you appear, where in the answer, and which competitors appear in your place. This baseline is the data you will be comparing against in Month 3.
Three topic cluster hubs. Pick your three highest-value topic areas and build or update a hub page for each, structured around the full cluster of sub-questions your target buyer might ask. Each hub should open with a definition-first sentence that directly answers the implied question of the page title, use clear H-tag hierarchy for every sub-question, and include at least one original data point, case reference, or expert quotation that is not available from a simple web search.
Author bio and E-E-A-T signals. Add or update author bylines on every hub page and service page. Author bio pages should list real credentials, years of experience, specific publications or recognitions where they exist, and a photo. This is often a one-day implementation task that has a disproportionate impact on AI trust-layer signals.
FAQ schema across top pages. For every page with a discernible question-and-answer structure, implement FAQPage schema. If the page does not have Q&A structure, add an FAQ section at the bottom addressing the top three implicit questions a reader of that page might still have.
Internal linking restructure. Audit your internal linking to ensure every hub page is linked to from at least 10-15 supporting pages, and that every supporting page links back to its hub. This is both a traditional SEO and an AI-retrieval improvement: it makes the relationship between your pages machine-readable as a topic cluster rather than a collection of disconnected documents.
Directory and listing push. Expand listing presence beyond the five core directories from Month 1 to the broadest relevant set for your industry. For local businesses this includes industry-specific directories and local chamber or association sites. For B2B and SaaS brands this includes G2, Capterra, Clutch, and relevant analyst-maintained lists. The goal is consistent entity information appearing across a wide surface area of the web.
Semrush's AI Visibility Index 2026 gives this a name worth adopting: the Citation Core, meaning the small set of third-party sources a given category's AI answers default to trusting, again and again, regardless of platform. Manage presence on those sources deliberately rather than treating third-party mentions as generic outreach. Based on the verticals we work in most, here's a starting Citation Core map:
| Vertical | Citation Core sources | Why it matters here |
| Local services & community orgs (faith, veterinary, home services) | Google Business Profile, Yelp, industry-specific directories (state vet boards, local chamber/association sites) | Google AI Mode over-indexes local-discovery brands like Yelp more than any other platform, making this the highest-leverage single fix for local clients. |
| Legal & immigration services | Avvo, Martindale-Hubbell, state bar directories, category-specific legal publications | High-trust, high-complexity categories where buyers research extensively; third-party credential corroboration matters more than volume. |
| Healthcare & consumer health | Healthline, WebMD, NIH.gov, and condition-specific authority sites and communities | AI Overviews leans hardest on established health authorities for this category; owned content alone won't carry it. |
| B2B software & SaaS | G2, Capterra, TrustRadius, relevant analyst lists | This is ChatGPT's and AI Mode's default citation layer for the category; Section 6's monitoring guidance already points here, this is the same logic applied to citation-building. |
| Finance & financial services | NerdWallet, Investopedia, Bankrate | Finance is the flattest category Semrush measured, with only ~41% of top-10 mentions held by the top three brands, meaning there's real room to move with focused, sustained presence. |
PR and third-party mention outreach. Pursue at least two or three earned mentions in third-party publications that are themselves visible in AI answers for your category. These do not need to be major national publications. Industry newsletters, professional association blogs, and niche vertical publications that your target audience reads are often both more achievable and more valuable for AI-entity corroboration than broader coverage.
Tool onboarding. If Month 1 and 2 have confirmed that AI visibility is a real opportunity worth tracking systematically, onboard a monitoring tool, starting with Peec AI or Otterly.ai for most teams, Profound for enterprise scale. Configure your tracked prompt set and set a weekly review cadence.
30-day citation review. Run your full prompt set again, compare against your Month 1 baseline, and document what moved and what did not. This comparison is the data you need to brief Month 4 and beyond: which topic clusters are now producing citations, which structural gaps remain, and where competitor displacement is happening that requires a competitive response.
These are the questions I receive most consistently from clients, prospects, and peers when the topic of AI search visibility comes up. The answers reflect my current working view, informed by the research cited throughout this guide and direct experience from the engagements described in Section 10.
Both statements are partly true, which is why the debate tends to run in circles. The underlying authority signals are substantially the same: credible, well-structured, frequently updated content, consistent entity presence across the web, and genuine expertise demonstrated through authorship and original perspective. Where it genuinely differs is in the structural requirements for machine readability (definition-first writing, FAQ schema, explicit topic cluster architecture), in the measurement approach (citation share rather than just keyword rankings), and in the specific technical requirements for AI crawler access that have no equivalent in traditional SEO. I would describe it as the same discipline with additional requirements rather than a separate discipline, which has the practical implication that you do not need to abandon your existing SEO program to pursue AI visibility. You need to extend it.
No, and this is one of the more important practical distinctions to understand. Ahrefs found that approximately 28% of ChatGPT’s most-cited pages have zero Google organic visibility, which means domain authority as traditionally measured does not transfer cleanly to AI citation authority. A page can be highly cited by AI engines despite being invisible in traditional search if it is clearly structured, directly answers a specific question, and has readable entity signals around it. Conversely, a high-authority domain can be consistently overlooked by AI engines if its content is buried in poorly structured prose, lacks schema markup, or has AI crawlers blocked. Domain authority helps but it is neither necessary nor sufficient.
The honest answer is that it depends almost entirely on which layer of the Authority Model you are starting from. Technical Foundation and Structure fixes (crawlability, robots.txt, basic schema) can produce measurable citation improvements in 4-8 weeks for brands that already have substantive content. Substance-layer work (topic cluster content, E-E-A-T implementation, original data) typically takes 3-6 months to compound into consistent citation share. Signal-layer work (directory expansion, earned mentions, entity corroboration) is a 6-12 month investment with compounding returns. The veterinary practice case study in Section 10 is the fastest result I have seen: 90 days to dominant local search presence. That timeline is possible when starting from a very low Foundation base with an existing real-world authority that just needed to be made digitally accessible.
This is a legitimate debate with real business implications beyond SEO, and I am not going to pretend there is a clean universal answer. From a pure AI visibility standpoint, blocking AI crawlers eliminates any possibility of being cited. If your concern is content licensing, revenue cannibalization, or competitive exposure, those are real concerns worth weighing against the visibility trade-off. What I would caution against is blocking AI crawlers by default without having made that choice deliberately, which is what happens when Cloudflare or another security layer changes its default configuration and teams do not notice. This is an active risk right now, not a historical one: Cloudflare's new crawler categories (Search, Agent, Training) take effect for mixed-use crawlers on ad-supported pages starting September 15, 2026, and because Cloudflare classifies Googlebot as a mixed-use crawler under that rule, a site with AI blocking already switched on could lose Google Search visibility along with AI crawler access unless someone reviews the settings before that date. If you have chosen to block AI crawlers for business reasons, that is a legitimate decision. If it is happening accidentally, that is a Foundation-layer problem to fix, and worth checking before mid-September regardless of where else this sits on your priority list.
The highest-leverage schema types I consistently see move the needle are: Organization or LocalBusiness (entity clarity at the Core layer), Article or WebPage with Author markup (E-E-A-T signaling at the Inner Trust layer), FAQPage (direct Q&A structure that maps cleanly onto how AI engines parse answer candidates), and BreadcrumbList (which helps AI systems understand where a page sits within your site’s topic architecture). Beyond these, HowTo schema for instructional content and Product or Service schema for commercial pages are high-value additions for relevant content types. I would focus implementation in that order rather than trying to apply every schema type simultaneously.
The llms.txt standard, which proposes a structured file in your site’s root directory that describes your brand and content to AI systems, has generated significant discussion but limited evidence of material impact so far. Google has explicitly stated it has no plans to use it. Anthropic’s Claude and some other systems acknowledge it but have not published documentation of how it affects retrieval weighting. My current position is that it is a low-effort task worth doing if someone in your team has 30 minutes available, not because the evidence for its impact is strong but because its downside risk is zero and the standard may become more widely adopted. It is categorically not a substitute for the Foundation, Structure, and Substance work described in this guide.
Google's own AI Search guidance, published in 2026, addresses this directly and the answer is no: rewriting pages specifically for AI consumption, as a distinct exercise from writing them well for people and for traditional search, produces no measurable advantage. This lines up with everything else in this guide. The Structure and Substance layers of the Authority Model (Section 4) describe what actually works: definition-first writing, clear heading hierarchy, schema markup, and genuine topical depth. That's the same content a human reader and an AI retrieval system both want. Maintaining a separate “AI version” of a page is wasted effort this guide would not recommend.
No, and Google's guidance is explicit on this point too: manufactured or paid-for brand mentions, the kind some GEO vendors sell as a shortcut, don't improve AI search visibility. This matches the mentions-versus-citations distinction in Section 2.6. Mentions are earned through brand authority and genuine category fit, not planted through paid placements, and AI systems appear reasonably resistant to that kind of manipulation, at least as of this writing. If a vendor is pitching mention injection as a service, treat it the same way you'd treat a link-farm pitch in traditional SEO.
This guide distinguishes throughout between claims supported by peer-reviewed or independently replicated research, claims sourced from a single vendor or proprietary study (noted as such), and claims that represent current industry inference or practitioner consensus without a strong empirical basis. All three types of claims appear in this field; the distinction matters for how much weight to place on any individual finding. Where a specific figure (such as a conversion rate multiple or a citation overlap percentage) varies across sources, I present the range rather than a single number and note the source disagreement.
Research conducted between January and August 2026 for primary sources; earlier academic research (KDD 2024) included on the basis of its peer-reviewed methodology and its status as the field’s primary empirical foundation.
Figures that appear as ranges (such as conversion rate multiples or citation overlap percentages) reflect genuine disagreement between independent sources rather than uncertainty on the part of this author. Where a single-source figure is presented without a range, it is because only one credible published source was available for that specific data point at time of writing, and this is noted in context.
This guide will be reviewed and updated as new research is published. Last updated: August 2026, refreshed with current ChatGPT and Google AI Overviews scale figures, the Ahrefs/BrightEdge data on declining AI Overview-to-top-10 citation overlap, Cloudflare's September 2026 mixed-crawler enforcement deadline, and current AI visibility tool pricing following the Scrunch AI acquisition by Sitecore and Profound's Series C. For the most current version and associated downloads, visit sanjayb.com/ai-authority.
© 2026 Sanjay B. Consulting | SanjayB.com | All rights reserved.